Neuro-control Method Research Articles

Neurocontrol methods in the context of development of technical solutions for transition to unmanned navigation

The paper considers the features and prospects of using neurocontrol methods in the context of development of technical solutions for transition to unmanned merchant vessels. The paper suggests a non-iterative training based artificial neural network (ANN), which is based on the principles of “direct inverse control” to control the speed and motion of unmanned surface vessels. The model is identified, and the structure of an artificial neural network and the diagram of the automatic control system (ACS) of an unmanned vessel (UV) are considered on the example of an electric propulsion vessel. A series of computational experiments is carried out to obtain a sufficiently complete training sample. and the control law is presented. The principle of the control system for an unmanned vessel is considered based on a neural network. At the next stage of the study, focus is on the synthesis of the optimal control system for UV navigation. The problem of the fastest motion of a third-order control object from one point (with any initial speed) to another (at the end point the vessel stops and the speed is zero) is considered. Based on the results of a series of experiments with the UV model, the controller parameters that provide the best indicators of control quality were set in the MATLAB Simulink environment.

JUSTIFICATION OF PARAMETERS OF THE CONTROL SYSTEM OF ELECTROMECHANICAL DEBALANCE VIBRODRIVE OF VIBRATION MACHINES ON THE BASIS OF ARTIFICIAL NEURAL NETWORK

The article analyzes and proposes an approach to the construction of a control system for electromechanical debalance vibrodrive for vibration machines based on an artificial neural network. As a result of the analysis of various methods of managing dynamic objects it is concluded that the most appropriate and perfect for this type of machine is neurocontrol method of predictive model neurocontrol, which allows to expand the functionality of these vibrating machines and significantly save energy for vibratory drive of their oscillations. A direct neuro-emulator is used to predict the future behavior of the oscillating mechanical system of the vibration technological machines and to calculate errors. An important feature of the predictive neurocontrol model in the proposed method of controlling the operation of vibrating technological machines using an artificial neural system is that there is no neurocontroller that needs to be trained, its place is taken by the optimization algorithm. Applying the proposed method of controlling operation of adaptive vibration technology machines using artificial neural network will optimize the electromechanical control of debalanced vibration drive of vibrating machines and provide optimal resonant modes of its operation (which is energy efficient) in all technological modes of vibrating operation. The technical and economic characteristics of this control method are further improved due to the fact that the proposed control method uses the technology of predictive model neurocontrol and as a result is constantly calculated (forecasted) several cycles in advance and determines the best strategy to control the frequency of forced cyclic vibration. As a result, the mechanical system of vibration machines spends less time in non-resonant mode. This method of control also minimizes the duration of transients when changing the load mass of the working body vibration or changing the mode of vibration parameters and the parameters of their technological process.

Restoring Behavior via Inverse Neurocontroller in a Lesioned Cortical Spiking Model Driving a Virtual Arm.

Neural stimulation can be used as a tool to elicit natural sensations or behaviors by modulating neural activity. This can be potentially used to mitigate the damage of brain lesions or neural disorders. However, in order to obtain the optimal stimulation sequences, it is necessary to develop neural control methods, for example by constructing an inverse model of the target system. For real brains, this can be very challenging, and often unfeasible, as it requires repeatedly stimulating the neural system to obtain enough probing data, and depends on an unwarranted assumption of stationarity. By contrast, detailed brain simulations may provide an alternative testbed for understanding the interactions between ongoing neural activity and external stimulation. Unlike real brains, the artificial system can be probed extensively and precisely, and detailed output information is readily available. Here we employed a spiking network model of sensorimotor cortex trained to drive a realistic virtual musculoskeletal arm to reach a target. The network was then perturbed, in order to simulate a lesion, by either silencing neurons or removing synaptic connections. All lesions led to significant behvaioral impairments during the reaching task. The remaining cells were then systematically probed with a set of single and multiple-cell stimulations, and results were used to build an inverse model of the neural system. The inverse model was constructed using a kernel adaptive filtering method, and was used to predict the neural stimulation pattern required to recover the pre-lesion neural activity. Applying the derived neurostimulation to the lesioned network improved the reaching behavior performance. This work proposes a novel neurocontrol method, and provides theoretical groundwork on the use biomimetic brain models to develop and evaluate neurocontrollers that restore the function of damaged brain regions and the corresponding motor behaviors.

Data-based Suboptimal Neuro-control Design with Reinforcement Learning for Dissipative Spatially Distributed Processes

For many real complicated industrial processes, the accurate system model is often unavailable. In this paper, we consider the partially unknown spatially distributed processes (SDPs) which are described by general highly dissipative nonlinear partial differential equations (PDEs) and develop a data-based adaptive suboptimal neuro-control method by introducing the thought of reinforcement learning (RL). First, based on the empirical eigenfunctions computed with Karhunen–Loève decomposition, singular perturbation theory is used to derive a reduced-order model of an ordinary differential equation that represents the dominant dynamics of the SDP. Second, the Hamilton–Jacobi–Bellman (HJB) approach is used for the suboptimal control design, and the thought of policy iteration (PI) is introduced for online learning of the solution of the HJB equation, and its convergence is established. Third, a neural network (NN) is employed to approximate the cost function in the PI procedure, and a NN weight tuning algorithm based on the gradient descent method is proposed. We prove that the developed online adaptive suboptimal neuro-controller can ensure that the original closed-loop PDE system is semiglobally uniformly ultimately bounded. Finally, the developed data-based control method is applied to a nonlinear diffusion-reaction process, and the achieved results demonstrate its effectiveness.

Neuro-PID Control for Electric Vehicle

In this paper we consider the neuro-control method and its application to control problems of an electric vehicle. The neuro-control methods adopted here is based on Proportional-plus-Integral-plus-Derivative (PID) control, which has been adopted to solve process control or intelligent control problems. In Japan about eighty four percent of the process industries have used the PID control. After deriving the self-tuning PID control scheme (neuro-PID) using the learning ability of the neural network, we will show the control results by using the speed and torque control of an electric vehicle.

Optimal Neurocontroller for Nonlinear Benchmark Structure

A neurocontrol method is applied to the nonlinear benchmark control problem. A neurocontroller is trained based on a reduced-order linear design model, then it is used to control a nonlinear evaluation model. In training the controller, a sensitivity evaluation scheme is used and weights are updated by minimizing a cost function. Absolute accelerations directly measured from sensors are used as the feedback signals for the controller. Not only the current step acceleration, but delay signals of sensor readings, are used to enhance the training capability. Numerical examples show that the controlled responses are considerably reduced compared with the uncontrolled case. In conclusion, the possibility of the proposed control algorithm as a candidate for the controller of nonlinear building is shown.

A mixture density network approach to modelling and exploiting uncertainty in nonlinear control problems

In nonlinear and stochastic control problems, learning an efficient feed-forward controller is not amenable to conventional neurocontrol methods. For these approaches, estimating and then incorporating uncertainty in the controller and feed-forward models can produce more robust control results. Here, we introduce a novel inversion-based neurocontroller for solving control problems involving uncertain nonlinear systems which could also compensate for multi-valued systems. The approach uses recent developments in neural networks, especially in the context of modelling statistical distributions, which are applied to forward and inverse plant models. Provided that certain conditions are met, an estimate of the intrinsic uncertainty for the outputs of neural networks can be obtained using the statistical properties of networks. More generally, multicomponent distributions can be modelled by the mixture density network. Based on importance sampling from these distributions a novel robust inverse control approach is obtained. This importance sampling provides a structured and principled approach to constrain the complexity of the search space for the ideal control law. The developed methodology circumvents the dynamic programming problem by using the predicted neural network uncertainty to localise the possible control solutions to consider. A nonlinear multi-variable system with different delays between the input–output pairs is used to demonstrate the successful application of the developed control algorithm. The proposed method is suitable for redundant control systems and allows us to model strongly non-Gaussian distributions of control signal as well as processes with hysteresis.

Neuro-control of cable vibration using semi-active magneto-rheological dampers

A neuro-control method is proposed for semi-active vibration control of stay cables using magneto-rheological (MR) dampers. A finite element model with multiple degrees-of-freedom is formulated for the hybrid system of a sagged stay cable incorporated with MR dampers. Two neural network control strategies are developed by using the full-order system model directly and introducing a reduced-order modal model, respectively. In the first strategy, which eschews a model reduction process, the neural network controller is designed and trained to have the dual functions of controller and observer, so that the resulting neuro-controller is able to perform the clipped linear quadratic Gaussian (LQG) control with incomplete state observation. In the second strategy, with the aid of a reduced-order modal model, the neural network controller is designed to consist of a control network and a state estimator network. The resulting neuro-controller performs the clipped LQG control using only a few observed states. The control effectiveness of the proposed neural network control strategies is verified numerically by application to a 12 m long scale-model cable prototype transversely connected with an MR damper near the lower anchorage. The response mitigation ability of the proposed control strategies using incomplete output observation is compared with that of the clipped LQG control with full state observation. The adaptive capability of the configured neuro-controllers under dynamic excitation distinct from training loading is examined. The analysis results show that the proposed control strategies can effectively implement semi-active vibration control of stay cables with the use of MR dampers.

Neural networks for structural control of a benchmark problem, active tendon system

Methodology for active structural control using neural networks has been proposed by Ghaboussi and his co-workers 1-8 in the past several years. The control algorithm in the mathematically formulated methods is replaced by a neural network controller (neuro-controller). Neuro-controllers have been developed and applied in linear and nonlinear structural control. Neuro-controllers are trained with the aid of the emulator neural networks. The emulator neural network is trained to learn the transfer function between the actuator signal and the sensor reading and it uses that past values of these quantities to predict the future values of the sensor readings. In this paper, we apply the previously developed neuro-control method in the benchmark problem of the active tendon system. The emulator neural network is developed and trained using the evaluation model given in the benchmark problem which is considered to be the true representation of the active tendon system. However, a reduced-order model has been developed and used, along with the emulator neural network, to train the neuro-controller. The evaluation model represents the three story steel frame structure, including the actuator dynamics. The absolute acceleration of the first floor and the actuator piston displacement are used as feedback. Three neuro-controllers, with different control criteria, have been developed and their performances have been evaluated with the prescribed performances indexes. The robustness of the neuro-controllers in the presence of some severe uncertainties, has also been evaluated.

Neural networks for structural control of a benchmark problem, active tendon system

Methodology for active structural control using neural networks has been proposed by Ghaboussi and his co-workers in the past several years. The control algorithm in the mathematically formulated methods is replaced by a neural network controller (neuro-controller). Neuro-controllers have been developed and applied in linear and nonlinear structural control. Neuro-controllers are trained with the aid of the emulator neural networks. The emulator neural network is trained to learn the transfer function between the actuator signal and the sensor reading and it uses that past values of these quantities to predict the future values of the sensor readings. In this paper, we apply the previously developed neuro-control method in the benchmark problem of the active tendon system. The emulator neural network is developed and trained using the evaluation model given in the benchmark problem which is considered to be the true representation of the active tendon system. However, a reduced-order model has been developed and used, along with the emulator neural network, to train the neuro-controller. The evaluation model represents the three story steel frame structure, including the actuator dynamics. The absolute acceleration of the first floor and the actuator piston displacement are used as feedback. Three neuro-controllers, with different control criteria, have been developed and their performances have been evaluated with the prescribed performances indexes. The robustness of the neuro-controllers in the presence of some severe uncertainties, has also been evaluated. © 1998 John Wiley & Sons, Ltd.

Nonlinear Structural Control Using Neural Networks

Recently, Ghaboussi and Joghataie presented a structural control method using neural networks, in which a neurocontroller was developed and applied for linear structural control when the response of the structure remained within the linearly elastic range. One of the advantages of the neural networks is that they can learn nonlinear as well as linear control tasks. In this paper, we study the application of the previously developed neurocontrol method in nonlinear structural control problems. First, we study the capabilities of the linearly trained neurocontrollers in nonlinear structural control. Next, we train a neurocontroller on the nonlinear data and study its capabilities. These studies are done through numerical simulations, on models of a three-story steel frame structure. The control is implemented through an actuator and tendon system in the first floor. The sensor is assumed to be a single accelerometer on the first floor. The acceleration of the first floor as well as the ground acceleration are used as feedback. In the numerical simulations we have considered the actuator dynamics and used a coupled model of the actuator-structure system. A realistic sampling period and an inherent time delay in the control loop have been used.

Consistency of HDP applied to a simple reinforcement learning problem

In “reinforcement learning over time,” a system of neural networks learns to control a system of motors or muscles so as to maximize some measure of performance or reinforcement in the future. Two architectures or designs are now widely used to address this problem in an engineering context: backpropagation through time and the adaptive critic family. This article begins with a brief review of these and other neurocontrol methods and their applications. Then it addresses the issue of consistency in using Heuristic Dynamic Programming (HDP), a procedure for adapting a “critic” neural network, closely related to Sutton's method of temporal differences. In a multivariate linear environment, HDP converges to the correct system of weights. However, a variant of HDP—which appeals to common sense and which uses backpropagation with a complete gradient—leads to the wrong weights almost always. Similar consistency tests may be useful in evaluating architectures for neural nets to identify or emulate dynamic systems.